Variable Construction

The dependent variable used in regressions at the empirical part of this thesis is ΔCash, which is the difference between cash at the ending of the year (t) and cash at the beginning of the year (t-1) scaled by book value of total assets at the beginning of the year (t-1). Main explanatory variables can be divided to two groups: cash sources and precautionary motive proxies. The construction of these variables is discussed in next two sections. All variables used in empirical regressions in Section 6 are generated following the methods used by McLean (2011). Furthermore, if some data values (such as R&D expenditures or other income that are not reported by all companies) are missing from companies that are active in that particular year, these values are consistently assumed to be zero.

4.2.1. Cash Sources

A company can have both internal and external sources of cash and even this kind of simple split between cash sources could be used in order to examine their effect on changes in cash ratios. However, internal cash sources can be further divided to operational and non- operational cash flows. Similarly, external cash sources can be divided to equity and debt issuances.

Issue is an item in cash flow statement and it represents cash proceeds from equity sales. It is the amount of euros received from share issuances during the financial year, scaled by lagged total assets. Thus, it does not distinguish between different types of equity issuance proceeds but all issuances are included as long as they create cash flow for the company. For instance, mergers financed with stock are excluded as they do not result cash proceeds. Because Issue is scaled by total assets at the beginning of the year, the sample does not include any cash proceeds from initial public offerings (IPOs) due to a technical reason. For example, if a company was publicly listed (i.e. it arranged an IPO) during 1995, its issue proceeds should have been scaled by assets at the end of 1994. However, Thomson ONE Banker reports data items only since the company has become public and therefore there would be no total assets reported for the company at the end of 1994. On the other hand, Issue is not limited to seasoned equity offerings only but it includes also any other equity sale that results for a cash

proceed to the company. For this reason, different kinds of equity sales are not distinguished but all share issuance proceeds are treated similarly in the scope of this thesis.

Debt is cash proceeds from debt sales scaled by lagged total assets. Thus, there is no difference whether the issued debt is short-term or long-term in nature. It is derived from balance sheet as the difference of total debt at the end of year and total debt at the beginning of year. As Debt represents specifically cash inflows for the company, it should not have a negative value and therefore all negative differences are marked as zero, indicating that the company has not made debt sales during the year. Debt sales as a cash flow statement items were largely missing in Thomson ONE Banker and that is why the variable is constructed using balance sheet items. Moreover, this method is a simple way to include increase of all kinds of debts: whether it is an increase of short-term credit line or long-term debt issuances.

Cash flow is derived from income statement as net income plus depreciation and amortization, scaled by lagged total assets. Thus, all internally generated operational turnover is not classified as cash flow because (usually) a majority of this income is not available for free use for the company but large part of turnover is used to cover different kinds of costs that generate the income. In other words, cash flow in this context means the amount of internally generated cash that is the result of company’s operations, i.e. net income. Depreciation and amortization are added to net income because they do not have real effect on cash flow but their effect on net income is derived from balance sheet. There are also other manners to construct the cash flow variable. For example, Bates, Kahle and Stulz (2009) define it as EBITDA minus interest, taxes, and common dividends. However, my definition for internal cash flow follows the one by McLean (2011).

Other represents all other cash sources that are not included in Issue, Debt, or Cash flow.

Thus, it basically includes cash inflows from sales of investments and sales of plant, property and equipment. Other is reported as income statement figure “other income” in Thomson ONE Banker and scaled by lagged total assets. Due to its nature, Other is more extraordinary cash source than other three cash sources. It includes cash inflows that are received from non- operational business transactions that don’t usually occur every financial year.

4.2.2. Precautionary Motive Proxies

Financial literature has agreed on three proxies that are able to capture the existence and scope of precautionary motives within individual companies: industry cash flow volatility, R&D expenses and dividends (see e.g. Opler et al. 1999, and Bates, Kahle and Stulz, 2009). In addition, McLean (2011) has created an index of the three above mentioned proxies in order to capture the precautionary motives of individual company to a one index called PREC.

Following McLean, I also create four different proxies to measure precautionary motives of sample firms.

Cash flow volatility is the natural logarithm of industry cash flow standard deviation (cash flow is defined in Table 2). First, natural logarithm of cash flow volatility over the last five years is calculated for each individual company for each sample year, a minimum of three observations is required. Then, outliers are excluded by winsorizing at 1% level. Next, companies are divided to industries by first two digits of their SIC codes. Finally, yearly industry cash flow volatility is retrieved by taking the average of industry firms’ cash flow volatility within two-digit SIC code industry classes. The reasoning of using cash flow volatility as a precautionary motive proxy is that companies within industries that have more unreliable cash flows (i.e. higher Cash flow volatility) tend to hold higher amounts of cash in order to be prepared for low cash flows during bad years.

Dividends is paid cash dividends scaled by lagged total assets. To notify, rationale for using dividends as precautionary motive proxy is not all straightforward. Bates, Kahle and Stulz (2009) conclude that non-dividend payers hold more cash than dividend payers and that their cash ratios have been increasing recently. Moreover, Fazzari et al. (1988) and Han and Qui (2007) state that firms that do not pay dividends are financially more constraint than dividend payers and that is the reason why they hold higher precautionary cash savings. On the other hand, reason not to pay dividends might occur also if a firm is growing fast and needs to have precautionary cash savings in order to make new investments whenever appropriate. Thus, the decision not to pay dividends is not automatically related to financial constraints, but more on future prospects. Even though many papers have supported the use of dividends as precautionary motive proxy, McLean (2011) treats dividends with caution. This is because the relation between dividends and cash holdings can exist also mechanically: if a firm decides not to pay dividends, then it will have more cash compared to decision to pay dividends, all else equal (McLean, 2011). Thus, Dividends is a proxy that needs to be interpreted carefully

while its explanatory power in relation to cash holdings might be ambiguous, but at the same time it is also interesting to compare findings in empirical part to recent studies that are made in U.S. context.

R&D is research and development costs scaled by lagged total assets. Firms that spend more on R&D are observed to hold higher levels of cash (Opler et al. 1999, and Bates, Kahle and Stulz, 2009). This is because R&D-intense firms have usually more valuable investment opportunities on sight and that is why they need to be prepared to utilize them by keeping precautionary cash holdings. As argued in case of Dividends, McLean (2011) again points out the obvious: R&D actions spend cash and therefore R&D and cash holdings might have negative relation as well. However, both Opler et al. and Bates, Kahle and Stulz have shown that generally R&D spending is associated with higher cash holdings and that R&D as precautionary motive proxy is well justified. As it is noted in studies executed in U.S. context, most companies don’t report any R&D expenses during financial year. Same lack of data is present for European firms in Thomson ONE Banker and therefore majority of R&D observations are forced to be marked as zero.

Following McLean (2011), I construct one additional precautionary motive proxy called PREC from three above-mentioned precautionary motive proxies. PREC is the first principal component of Cash flow volatility, Dividends and R&D. In other words, each of three proxies is likely to contain both precautionary motives component and component that is not connected to precautionary motives. PREC is thus created in order to capture the common precautionary component in each of these proxies (see e.g. Jolliffe, 2005⁷). Due to the nature of proxies discussed in earlier paragraphs, PREC is expected to be higher for firms with high industry cash flow volatility, low-dividend payers and firms with high R&D spending. PREC is also constructed using only Cash flow volatility and R&D due to the ambiguous interpretation of Dividends as a precautionary motive. This method however results to similar findings compared to PREC where Dividends is included⁸. Therefore, I decided to report only the results using the PREC that is constructed using all three precautionary motive proxies.

PREC is calculated for the whole sample (i.e. all firm-year observations) at one time in order to make the first principal component comparable for each year and each firm. In order to a single firm to retrieve a value for PREC, it needs to have observation for all its components.

7Brief definition of principal component analysis is stated e.g. in Jolliffe (2005): “The central idea of principal component analysis is to reduce the dimensionality of a data set consisting of a large number of interrelated variables, while retaining as much possible of the variation present in the data set.”

8 These two alternative methods for constructing the first principal component have a correlation of over 0.800.

In other words, if observation for Cash flow volatility, R&D, or Dividends is missing, PREC cannot be calculated. However, as mentioned earlier, if firm does not report e.g. R&D during a financial year, it is assumed to be zero. Otherwise the limited availability of R&D observations would dramatically decrease the amount of observations for PREC as well. The construction of PREC results for first principal components (or eigenvectors) of 0.701, 0.619 and -0.354 for Cash flow volatility, R&D and Dividends, respectively. Thus, signs for eigenvectors are as expected as increase in Cash flow volatility and R&D have positive effect on PREC (positive components) and Dividends has negative effect (negative component).